The MOS Silicon Gate Technology and the First Microprocessors
Total Page：16
File Type：pdf, Size：1020Kb
The MOS Silicon Gate Technology and the First Microprocessors Federico Faggin This is a preprint of the article published in La Rivista del Nuovo Cimento, Società Italiana di Fisica, Vol. 38, No. 12, 2015 1. – Introduction There are a few key technological inventions in human history that have come to characterize an era. For example, the animal-pulled plow was the invention that enabled efficient agriculture, thus gradually ending nomadic culture and creating a new social order that in time produced another seminal invention. This was the steam engine which gave rise to the industrial revolution and created the environment out of which the electronic computer emerged – the third seminal invention that started the information revolution that is now defining our time. All such inventions have deep roots and a long evolution. Engines powered by water or wind were used for many centuries before being replaced by steam engines. Steam engines were then replaced by internal combustion engines, and finally electric motors became prevalent; each generation of engines being more powerful, more efficient, more versatile, and more convenient than the preceding one. Similarly, the origin of our present computers dates back to the abacus, a computational tool that was used for several millennia before being replaced by mechanical calculators in the 19th century, by electronic computers in the 1950’s, and by microchips toward the end of the 20th century. The invention of the electronic computer was originally motivated by the need for a much faster computational tool than was possible with electromechanical calculators operated by human beings. This improvement was accomplished not only by performing the four elementary operations considerably faster than previously possible with calculators, but even more importantly, by adding the ability to program a long sequence of arithmetic operations that could be executed automatically, without human intervention. The addition of programmability, proved to be an immensely fruitful and versatile capability, allowing computers to soon become universal symbol manipulators, with countless new applications, far exceeding what even programmable calculators were originally intended to do. Toward the end of the 20th century, the relentless progress in microelectronics combined with the digitization of all types of information, made computers so powerful, small, low-cost, and pervasive that many functions that previously required separate, single-function devices were subsumed by a single programmable mobile computer capable of handling the individual’s needs for communication, computation, control, and storage of all kinds of information, be it numbers, text, images, sound, or video. This paper will describe the history of two key inventions: the MOS silicon gate technology, and the microprocessor. These were the two developments that made it possible to replace huge and costly machines with pocket-size devices many thousands of times less expensive, and thousands of times more powerful than the early computers. This remarkable progress was due to the power and flexibility of microelectronics, the technology that provided the early transistors used in the second-generation computers. What’s remarkable is that while in the late 1950’s transistors were just one of the many key components necessary to build a computer, less than 20 years later an entire monolithic computer could be built in a single chip of silicon, in the same physical volume that previously housed a single transistor! This progress is unprecedented because in less than 30 years, a single chip weighting less than one gram, occupying a volume smaller than a cubic centimeter, dissipating less than one Watt, and selling for less than ten dollars could do more information processing than the UNIVAC I, the first commercial computer, which used 5200 vacuum tubes, dissipated 125 kW, weighted 13 metric tons, occupied more than 35 m2 of space, and sold for more than one million dollars per unit. Viewed from today’s perspective, the early giant computers of the 1950’s and ‘60’s provided precisely the architectural blueprints of the type of symbol manipulator that the world needed. The microelectronics industry then did the rest, bringing the cost, size, and energy requirement down to the point where a computer could fit inside an electric toothbrush, a hearing aid, or an inexpensive toy – applications that were not only unimaginable, but even incongruous when computers were room-sized and cost millions of dollars. After a brief history of microelectronics and computers, this paper will describe, from a first-person basis, the development of the silicon gate technology (SGT) at Fairchild Semiconductor, and the development of the early microprocessors (MP) at Intel, the two inventions that gave new life and impetus to the information revolution that began in the mid 1940’s with the development of the first mainframe electronic computers using vacuum tubes. 2. – A brief history of microelectronics Microelectronics officially started with the invention of the first transistor at Bell Laboratories in 1947 by John Bardeen, Walter Brattain and William Shockley, just one year after the first electronic computer, the ENIAC, was realized at the University of Pennsylvania. No one ever imagined at that time that these two inventions would merge, less than 30 years later, into a microchip, changing society in a fundamental way. By replacing the electro-mechanical relays used in previous generations of calculating machines with vacuum tubes, the operating speed of ENIAC was increased by more than a factor of 1,000. This increase in speed came at a cost — vacuum tubes were bulky, power-hungry, and most of all, unreliable. The short meantime between failures in a single vacuum tube was a crippling problem for a computer needing thousands of them. Considering the reliability issues that then plagued electronic equipment, vacuum tube computers were over 10 times more likely to malfunction when compared to the most complex electronic equipment of that time. The search for a viable vacuum tube replacement with a more reliable, smaller, power-efficient, and lower cost solid state device had been ongoing since the 1920’s. The first patent for a semiconductor “triode” was filed in 1925 by Julius Edgar Lilienfeld, and a more advanced device was patented by Oskar Heil in 1934. Both devices were field-effect devices, similar in principle to the current MOS transistors, but no commercial devices were ever produced. The first Bell Labs transistor was a point-contact device, commercialized in 1948 by Raytheon with model CK703. However, point-contact transistors were too difficult to build and too fragile to be useful, since even a modest mechanical shock could put an end to their operation. A new operating principle was needed. It was the breakthrough work of W. Shockley on diffusion transistors that introduced the new operating principles used in all modern bipolar transistors. This original work led to the first commercial alloy- junction transistors introduced by GE and RCA in 1951. From that point on, new applications for bipolar transistors started to grow rapidly – among them, the first hearing aids and portable radios – and in 1953, one million transistors were produced in the US alone. All early transistors used a tiny single-crystal of germanium as their starting semiconductor material, and were built one at a time, just like the vacuum tubes. It was soon realized, however, that a better semiconductor material was necessary, since the temperature effects on germanium transistors could cause a thermal runaway with the self-destruction of the transistor. This unwanted effect was due to the rather narrow bandgap (.66 eV at 300K) between the conduction band and the valence band of germanium crystals. Furthermore, the operating frequency of germanium transistors was rather limited. A better material, silicon, with a bandgap of 1.11 eV at 300K, was soon identified, and silicon junction transistors were introduced commercially for the first time by Texas Instruments (TI) in 1954. In 1957, a new company, Fairchild Semiconductor, was started in the San Francisco Bay Area by eight key engineers abruptly leaving Shockley Semiconductor. Among them were Robert Noyce, Gordon Moore, Jean Hoerni, and Jay Last. Fairchild’s mission was to develop advanced bipolar junction transistors made with silicon to serve the needs of the emergent aerospace industry. Just for size, in 1957 the total production of transistors in the US was 29 million units. Today, a single chip costing less than $1 may easily contain more than 30 million transistors, including all their interconnections. Before long, Fairchild Semiconductor became the leading company in the nascent microelectronics industry due to the seminal invention of the planar process by Jean Hoerni, a Swiss engineer. Up until that time, transistors had been fabricated one at a time. With the planar process, many transistors could be simultaneously fabricated, one next to the other, on the surface of a thin slice (called wafer) of a single-crystal silicon ingot. The planar process truly revolutionized the industry because it not only dramatically reduced the size and cost of transistors, but far more importantly, it made possible the monolithic integrated circuit (IC). Hoerni’s process consisted in fabricating an array of identical transistors on the surface of a silicon wafer with the diameter of less than one inch. This was done by recognizing that silicon dioxide could mask the diffusion of dopants in silicon. Therefore, the silicon doping necessary to create semiconductor junctions in specific places, could be achieved by first thermally growing a layer of silicon dioxide on the surface of a silicon wafer, and then opening windows in the oxide where the junctions were needed. The windows were defined by using photolithography followed by etching, a technique already in use to make printed circuit boards. After the removal of the oxide by chemical etching in the areas not protected by the developed photoresist, the junctions could then be created by thermal diffusion of the appropriate dopants into the silicon.